放射组学鉴别破裂颅内动脉瘤:概述、基于METRICS和RQS的方法学质量评价及独立多中心数据集的特征可迁移性验证
系统综述26项鉴别破裂颅内动脉瘤的放射组学研究的质量,平均RQS为9.7(SD 5.2),METRICS中位分72.5%(IQR 58.5%-76.2%);在MIRACLE队列(5中心)验证15项研究的特征集,外部测试AUC为0.51-0.61,表明特征可迁移性中等。
要点
- 外部测试集中报告的特征集AUC仅为0.51-0.61,多数特征可迁移性有限。
- original_shape_VoxelVolume、MeshVolume和SurfaceArea在自动与手动分割间ICC>0.9,稳定性最高。
核心结论
- 鉴别破裂颅内动脉瘤的放射组学研究方法学不断发展但异质,部分特征稳定性高但可迁移性中等,方法学缺陷包括外部验证有限、模型泛化性欠佳和开放科学依从性不足。
研究设计与样本
- 系统综述放射组学文章,使用RQS和METRICS进行质量评价;在MIRACLE队列(5个中心,自动分割和特征提取)中评估已识别特征集的可迁移性,并评估内部性能。
诊断标准与关键数值
- 26项研究的平均RQS为9.7(SD 5.2),METRICS中位分为72.5%(IQR 58.5%-76.2%)。
- 主要可迁移性分析中,训练数据集AUC范围为0.59至0.70,外部测试数据集AUC范围为0.51至0.61。
- original_shape_Elongation是最常使用的特征(15项研究中9项使用)。
- 比较手动与自动分割的放射组学特征,original_shape_VoxelVolume、MeshVolume和SurfaceArea稳定性最高,ICC均>0.9。
- 单项研究最高RQS为18,最高METRICS为84.6%。
临床应用与阅片要点
- 原文结论指出,未来研究应解决外部验证不足和透明度有限等方法学缺陷,以提升研究质量和临床相关性。
局限性
- 原文结论指出方法学缺陷包括外部验证有限、模型泛化性欠佳和开放科学依从性不足。
- 原文结论指出未来研究应解决这些缺陷以提升研究质量和临床相关性。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
英文原文(摘要)· 点击展开
OBJECTIVES: To assess the methodological quality of radiomics research for ruptured intracranial aneurysms (IAs) using the radiomics quality score (RQS) and METhodological radiomICs score (METRICS), and to evaluate the transportability of reported radiomics features using an independent multi-center dataset.
MATERIALS AND METHODS: We systematically reviewed radiomics articles for ruptured IAs classification. Included articles underwent a quality assessment via RQS and METRICS. Then, the identified radiomics feature sets were evaluated in the MIRACLE Cohort, which comprises IA cases from five centers with automated segmentation and feature extraction. The internal performance was assessed for each of the validated feature sets.
RESULTS: After screening 252 studies, 26 were analyzed, yielding a mean RQS of 9.7 (SD, 5.2) and a median METRICS score of 72.5% (IQR, 58.5%-76.2%). Two individual articles reached a maximum RQS of 18 and a METRICS of 84.6%, respectively. Fifteen studies were selected for transportability evaluation. In the primary transportability analysis, the AUCs ranged from 0.59 to 0.70 in the training dataset, and from 0.51 to 0.61 in the external testing dataset. The original_shape_Elongation feature was the most frequently utilized feature (9/15). When comparing radiomics features derived from manual and automatic segmentations, original_shape_VoxelVolume, MeshVolume, and SurfaceArea showed the highest stability, all achieving an intraclass correlation coefficient (ICC) > 0.9.
CONCLUSIONS: Radiomics studies for distinguishing ruptured IAs exhibit an evolving but heterogeneous methodology, with certain features demonstrating high stability but modest transportability. Methodological deficiencies remain, including limited external validation, suboptimal model generalizability, and insufficient adherence to open science. Future research should address these gaps to enhance study quality and clinical relevance.
TRIAL REGISTRATION: MIRACLE Cohort, ChiCTR2400084601. Registered 21 May 2024, https://www.chictr.org.cn/showprojEN.html?proj=229047 .
KEY POINTS: Question To evaluate the methodological quality of radiomics research for IA rupture using RQS/METRICS and assess feature transportability using independent multi-center data. Findings Methodological heterogeneities persist despite field growth. Independent assessment reveals that most reported radiomics features demonstrate modest transportability, with performance dropping across different centers. Clinical relevance Addressing methodological gaps, specifically insufficient external validation and limited transparency, is essential to facilitate the reliable translation of radiomics from academic research into clinical applications for aneurysm management.
引用
- Zhu D, Xu M, Wang J, et al. Radiomics for differentiating ruptured intracranial aneurysms: overview, methodological quality evaluation using METRICS and RQS, and feature transportability validation using independent multi-center dataset. Eur Radiol. 2026. PMID: 41925835. doi:10.1007/s00330-026-12485-9
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。